Model complexity and data requirements in snow hydrology: seeking a balance in practical applications

We investigate the problem of balancing model complexity and input data requirements in snow hydrology. For this purpose, we analyze the performance of two models of different complexity in estimating variables of interest in snow hydrology applications. These are snow depth, bulk snow density, snow water equivalent and snowmelt run-off. We quantify the differences between data and model prediction using 18years of measurements from an experimental site in the French Alps (Col de Porte, 1325m AMSL). The models involved in this comparison are a one-layer temperature-index model (HyS) and a multilayer model (Crocus). Results show that the expected loss in performance in the one-layer temperature-index model with respect to the multilayer model is low when considering snow depth, snow water equivalent and bulk snow density. As for run-off, the comparison returns less clear indications for identification of a balance. In particular, differences between the models' prediction and data with an hourly resolution are higher when considering the Crocus model than the HyS model. However, Crocus is better at reproducing sub-daily cycles in this variable. In terms of daily run-off, the multilayer physically based model seems to be a better choice, while results in terms of cumulative run-off are comparable. The better reproduction of daily and sub-daily variability of run-off suggests that use of the multilayer model may be preferable for this purpose. Variation in performance is discussed as a function of both the calibration solution chosen and the time of year. Copyright (c) 2016 John Wiley & Sons, Ltd.